【プロンプト】AIの回答精度を劇的に高める構造化プロンプトの型 by Prompt-Lab
再帰的思考プロセスを実装した、高精度な推論を導くためのプロンプトフレームワーク。
### CORE-FRAMEWORK: The Recursive Chain of Thought (R-CoT) ```python # SYSTEM_DIRECTIVE: R-CoT_v4.0 # OBJECTIVE: To maximize inferential accuracy through structured cognitive recursion. class CognitiveProcessor: def __init__(self, task_input): self.raw_data = task_input self.state = "INITIALIZATION" def execute_chain(self): # STAGE 1: DECONSTRUCTION (The Atomization) # Identify core variables, implicit constraints, and hidden intent. # Ignore surface-level framing; extract the fundamental problem vector. # STAGE 2: EPISTEMIC MAPPING (The Anchor) # Establish domain-specific axioms. Define what is 'known' vs 'speculative'. # Prevent hallucination by isolating the uncertainty boundary. # STAGE 3: SYNTHETIC RECURSION (The Loop) # Iteration 1: Draft initial hypothesis. # Iteration 2: Critique hypothesis against constraints from Stage 1. # Iteration 3: Refine. Repeat until error-rate < threshold. # STAGE 4: OUTPUT_TRANSFORM (The Crystallization) # Format as actionable intelligence. Minimal verbosity. Maximum density. pass ``` ### PROMPT-TEMPLATE: The "Omni-Perspective" Crucible [INPUT_DATA] --- [CONTEXT_LIMITS] - Logic: Deductive only. - Tone: Clinical, objective, devoid of fluff. - Constraints: Maintain 100% adherence to explicit instructions. [THOUGHT_PROCESS] 1. [DECODE]: Analyze the prompt for latent ambiguity. If found, define the parameters before proceeding. 2. [SIMULATE]: Run three distinct cognitive models: - Model A (The Skeptic): Challenges every assumption. - Model B (The Pragmatist): Focuses on implementation efficiency. - Model C (The Architect): Focuses on structural integrity and future-proofing. 3. [SYNTHESIZE]: Integrate the outputs. Eliminate contradictions. 4. [OUTPUT]: Execute the final response strictly in [OUTPUT_FORMAT]. [OUTPUT_FORMAT] - Concept: [Concise Definition] - Strategy: [Step-by-step Execution] - Risk: [Potential Failure Point] - Validation: [Evidence/Metric] --- ### PHILOSOPHICAL ENGINE: The Boundary Questioning Protocol To enhance output quality, the AI must first define the 'Non-Answer'. Use the following logic to prune the output space: